Huu-Nghia Nguyen, Christian Bonnet Mobile Communications Department
Eurecom Institute
2229 Route des Crêtes, Sophia Antipolis, France Huu-Nghia.Nguyen@eurecom.fr, Christian.Bonnet@eurecom.fr
Abstract— The Always Best Connected (ABC) vision has always been about IP-based architecture and multiple simultaneous access technologies which can provide wireless bandwidth aggregation and load-balancing features in fully overlapped coverage areas. This paper presents an intelligent tunneling framework, referred to as virtual SCTP tunneling, which enables the ABC vision in Network Mobility (NEMO). The virtual SCTP tunneling framework supports tunnels with multi-homed endpoints and can multiplex packets to multiple radio interfaces on a per-packet basis for better performance. Besides, it reduces the tunneling overhead over radio interfaces in heavy-load situation thanks to the predictive packet bundling feature. We carry out the simulation in the multi-homing context under ns2 and observe different metrics, e.g. throughput and packet loss rate seen by the wireless network, and the average end-to-end delay seen by Local Fixed Nodes. The performance of the proposed framework is compared with that of the existing per- flow forwarding which uses flow binding and IP tunneling. It is shown that the virtual SCTP tunneling is advantageous over the existing per-flow forwarding, especially in heavy-load situation.
Keywords- Always Best Connected, Wireless Bandwidth Aggregation, Load-balancing, Network Mobility, NEMO, Multi- homing, Virtual SCTP Tunneling. 4G.
I. INTRODUCTION
The Always Best Connected (ABC) concept [1] is considered as the vision beyond vertical handover between heterogeneous access technologies. Recent researches on vertical handover have taken advantage of multi-homing by using bi-casting in partially overlapped coverage areas to increase the handover performance. The ABC vision, in turn, consider multi-interface mobile nodes and multiple simultaneous access technologies in fully overlapped coverage areas to enable the simultaneous use of access technologies that is foreseen as a key feature of 4G.
Network Mobility (NEMO) Support [2] provides seamless mobility to Mobile Networks, which are defined as network segments or subnets that can move and attach to any points in the Internet topology. A Mobile Network includes one or more Mobile Routers (MRs) which connect it to the global Internet.
Nodes behind the Mobile Router, called Mobile Network
Nodes (MNNs), are Local Fixed Nodes (LFNs), Local Mobile Nodes (LMNs) and Visiting Mobile Nodes (VMNs). For the related terminology, see [3]. NEMO Basic Support [4]
describes protocol extensions to Mobile IPv6 [5] to enable support for network mobility. One advantage of NEMO Basic Support is that the Mobile Network Nodes need not be aware of the actual location and mobility of the mobile network.
With some approaches for Route Optimization [6], it might be necessary to reveal the point of attachment of the Mobile Router to the Mobile Network Nodes. This may mean a tradeoff between mobility transparency and Route Optimization.
This paper envisages NEMO with multi-interface Mobile Routers and with mobility transparency to Mobile Network Nodes. The biggest challenge toward the ABC vision is to allow multi-interface Mobile Routers to distribute the traffic simultaneously via different radio access technologies while moving to improve the performance in terms of throughput, packet loss rate and end-to-end delay. For simplicity, only Local Fixed Nodes are considered. All issues related to multi- homing in NEMO can be classified into two groups: the interaction between different entities to maintain multiple bindings simultaneously and the method of distributing the traffic simultaneously via multiple active radio interfaces. The first group of issues can be accomplished by the Multiple Care- of Addresses extension [7]. As for the second group of issues, per-flow forwarding approach is a solution for simultaneous access in NEMO and is still in the discussion of the IETF Monami6 Working Group. It is analyzed that, in NEMO, the per-flow forwarding approach using flow binding and IP tunneling requires more complexities for tunnel management, limits the granularity for simultaneous access to a per-flow basis, and can not be directly applied in NEMO without using a special mechanism for differentiating flows. We propose here a new intelligent tunneling framework, referred to as virtual SCTP tunneling, which allows Mobile Router traffic to be distributed via different active radio interfaces simultaneously on a per-packet basis and reduces the tunneling overhead over radio interfaces.
This paper is organized as follows. Section II provides a review on related work: the NEMO Basic Support, Multiple
An Intelligent Tunneling Framework for Always Best Connected Support
in Network Mobility (NEMO)
Care-of Addresses extension for Mobile IP, and per-flow forwarding with IP tunneling. Section III describes the new intelligent tunneling framework, i.e. virtual SCTP tunneling, for NEMO. Section IV provides performance evaluation on virtual SCTP tunneling and existing per-flow forwarding.
Finally, section V concludes the paper and provides perspectives for our future work.
II. RELATED WORK A. NEMO Basic Support
NEMO Basic Support [4] describes protocol extensions to Mobile IPv6 to enable support for network mobility. A Mobile Network is a network segment or subnet that can move and attach to any points in the Internet. A Mobile Network can only be accessed via Mobile Routers that manage its movement.
Mobile Networks have at least one Mobile Router serving them. A Mobile Router maintains a bi-directional tunnel to a Home Agent (HA) that advertises an aggregation of Mobile Networks to the infrastructure. A Mobile Router has a unique Home Address through which it is reachable when it is registered with its Home Agent. The Home Address is configured from a prefix aggregated and advertised by its Home Agent. The prefix could be either the prefix advertised on the home link or the prefix delegated to the Mobile Router.
When the Mobile Router has multiple interfaces or when there are multiple prefixes in the home link, The Mobile Router can have more than one Home Address.
When the Mobile Router moves away from the home link and attaches to a new access router, it acquires a Care-of Address from the visited link. As soon as the Mobile Router acquires a Care-of Address, it immediately sends a Binding Update to its Home Agent as described in [5]. When the Home Agent receives this Binding Update, it creates a cache entry that binds the Mobile Router's Home Address to its Care-of Address at the current point of attachment. The Mobile Router set a flag (R) in the Binding Update to indicate to the Home Agent that it acts as a Mobile Router and provides connectivity to nodes in the Mobile Network.
The extension defines a new Mobility Header Option for carrying prefix information. If the Mobile Network has more than one IPv6 prefix, it can include multiple prefix information options in a single Binding Update. The Home Agent sets up forwarding for each of these prefixes to the Mobile Router's Care-of Address and acknowledges the Binding Update by sending a Binding Acknowledgement to the Mobile Router.
Once the binding process finishes, a bi-directional tunnel is established between the Home Agent and the Mobile Router.
The tunnel end points are the Mobile Router's Care-of Address and the Home Agent's address. All traffic between the Local Fixed Nodes and Correspondent Nodes passes through the Home Agent and the tunnel. The Route Optimization is out of scope of NEMO Basic Support.
Even though NEMO Basic Support does not discuss multi- homing, [8] provides a full analysis on multi-homing in NEMO with different cases of multi-homing and related issues. In [9], an IPv6 soft handover extension for NEMO Basic Support (NEMO-SHO) with multi-interface Mobile Routers, using
packet bicasting and combining, has been proposed and experimented.
B. Multiple Care-of Addresses Registration
The objective of the IETF Monami6 Working Group (WG) [10] is to deal with the simultaneous use of multiple addresses for either Mobile Nodes using Mobile IPv6 or Mobile Routers using NEMO Basic Support. The Monami6 WG provides a protocol extension that supports the registration of multiple active IPv6 Care-of addresses for a given Home address [7] to allow the Mobile Node or the Mobile Router to get Internet access through multiple radio interfaces simultaneously. For doing so, a new identification number, called Binding Unique Identifier (BID), must be carried in each binding for the receiver to distinguish between the bindings corresponding to the same Home Address. The BID is used as a search key for a corresponding entry in the binding cache in addition to the Home Address. When a Home Agent and a Correspondent Node (CN) check the binding cache database for the Mobile Router, it searches a corresponding binding entry with the Home Address and BID of the desired binding. If necessary, a Mobile Router can use policy and filter information to look up the best binding per session, per flow, or per packet.
Note that a multi-interface Mobile Router can have more than one Home Address. Mobile IPv6 has mechanisms to manage multiple Home Addresses based on Home Agent's managed prefixes such as mobile prefix solicitation and mobile prefix advertisement. However those Home Addresses are seen as separated from each other. The Multiple Care-of Addresses Registration extension recommends assigning only a single Home Address to a Mobile Router with the assumption that applications will not need to be aware of the multiplicity of Home Addresses.
C. Per-flow Forwarding
Also in the Monami6 WG, the concepts of flow and flow binding are proposed: A flow is defined as one or more connections having the same flow identifier. A single connection is identified by the source and destination IP addresses, transport protocol number and the source and destination port numbers. A flow binding is a mobility binding extended with a flow identifier; it associates a particular flow to a Care-of Address without affecting other flows using the same Home Address.
An extension for flow binding [11] introduces the Flow Identifier Option, which is included in the Binding Update message and used to describe a flow to the recipient of the Binding Update. Using the Flow Identifier Option introduced in this specification a Mobile Node or Mobile Router can bind one or more flows to a Care-of Address while maintaining the reception of other flows on another Care-of Address. If the IP tunneling method and the flow binding are used to distribute flows via multiple active radio interfaces, the decision for flow binding must be done by Mobile Routers, based on local policies within the Mobile Router and based on information about network characteristics.
Supporting mobility transparency to Local Fixed Nodes implies that the flow binding policy exchange can not apply to
Local Fixed Nodes. This causes the lack of information of Local Fixed Nodes’ flows that is necessary for flow binding.
Besides, while Mobile Routers are moving from one wireless link to another, the dynamic nature of the network characteristic can also has negative impact on flow binding.
For these reasons, per-flow forwarding may become a rigid mechanism in NEMO. As a result, the performance (e.g.
throughput, end-to-end delay) for both the overall wireless network and the Mobile Network is limited. Besides, in a multi-homing context, IP tunneling supports only one address for each endpoint, the use of IP tunneling for multi-homing implies using a multitude of bi-directional tunnels and results certain complexities for the tunnel management at the Home Agent and Mobile Routers. In the next session, we present a new intelligent tunneling framework, called virtual SCTP tunneling, to overcome these limitations.
III. VIRTUAL SCTPTUNNELING FRAMEWORK
The virtual SCTP tunneling concept has been first presented in [12], and is based on a modification of Stream Control Transmission Protocol (SCTP) [13][14] and its extensions [15][16] that is designed to supports multi- streaming and multi-homing. The ‘virtual’ term signifies the fact that we apply the concepts of SCTP to tunnels having multi-homed endpoints. As a result, a virtual SCTP tunnel and a SCTP association are isomorphic, i.e. can be mapped onto each other, and we can reuse the design, and implementation of SCTP with minor modifications, principally in the encapsulating packet structure. The virtual SCTP tunneling is considered as a lightweight SCTP in terms of functionalities.
A. Conceptual Architecture
We consider tunnel having multi-homed endpoints as a virtual SCTP association between two virtual SCTP endpoints.
For a bi-directional tunnel, a pair of encapsulator and decapsulator must exist at each tunnel endpoint. The encapsulator is considered entry point of the tunnel, and the decapsulator is considered the exit point of the tunnel. In general, the same tunnel can be shared by different connections. In NEMO, one virtual SCTP endpoint is the Home Agent, and the other is the Mobile Router. Traffic from Correspondent Nodes to Local Fixed Node is tunneled between the Home Agent and the Mobile Router through the virtual SCTP tunnel (see Fig. 1).
For a normal encapsulation process, whenever an incoming packet is forwarded to the encapsulator, it will be encapsulated in an encapsulating packet which will later be delivered back to the IP routing layer. Instead of having packets distributed rigidly on a per-flow basis, in which each flow is mapped to a predefined radio interface, an intelligent scheduler is
introduced inside the encapsulator to allow the cooperation of different active radio interfaces and provides dynamic and flexible per-packet forwarding. A meter estimates the inter- arrival time of the incoming traffic, which is later used by the scheduler for the predictive packet bundling algorithm. On receiving an encapsulating packet, the decapsulator strips off the outer header, restores original encapsulated packets and delivers them back to the routing layer.
B. Encasulating Packet Structure and Packet Bundling The virtual SCTP tunneling method can bundle multiple small packets in one encapsulating datagram in case that multiple incoming packets are ready for processing in the tunnel’s queue. This technique allows all encapsulated packets between the Home Agent and the Mobile Router to share the same IP header and therefore reduces significantly the tunneling overhead over the radio interfaces. As there is no need to de-multiplex the traffic to particular applications at tunnel endpoints, the SCTP common header is eliminated to optimize the encapsulating datagram structure.
In the absence of packet bundling, the encapsulating datagram structure will be the same as in IP tunneling (IP-in-IP or GRE). Only one bit in the IP header is required to mark the presence of packet bundling; this can be the least significant bit (LSB) of the FlowID field in IPv6 or a reserved bit in IPv4. On presence of packet bundling, each incoming packet will be put in an encapsulating data chunk, of which the chunk header is the same as the 4-bytes SCTP common chunk header and has the form of Tag-Length-Value. Let k denote the number of encapsulated packets in one encapsulating packet; a value of k=1 implies the absence of packet bundling, a value of k>1 implies the presence of packet bundling. Fig. 2 shows a simple encapsulating datagram with k encapsulating chunks (k>1);
each encapsulating chunk contains an encapsulated packet. Let si be the size of the ith encapsulated packet. Let MTU, frameheader and IPheader respectively be the maximum transmission unit, the frame header size, and the IP header size.
The value of k must satisfy the following constraint to avoid the segmentation.
IPheader r
frameheade MTU
s k
ki
i
≤ − −
+ ∑
=1
4
C. Per-packet Dynamic Forwarding
In NEMO, the traffic between the Home Agent and a Mobile Router consists of many flows between different Correspondent Nodes and different Local Fixed Nodes. If the
Figure 1. A bi-directional virtual SCTP tunnel
Figure 2. An encapsulating datagram
Mobile Router doesn’t have any special mechanism to differentiate flows, then the traffic is seen as one big flow and should be distributed through different radio interfaces on a per-packet basis to increase the Mobile Network performance and the whole wireless network performance thanks to the wireless bandwidth aggregation. Even if the Mobile Router has a mechanism to differentiate flows, per-packet forwarding provides a better flexibility to increase the load-balancing performance.
D. Predictive Packet Bundling
The decision for packet bundling is basically done based on the tunnel’s queue length and/or in a predictive manner. The meter measures the inter-arrival time of the incoming stream of packets. Let t denote the elapsed time between two arrivals, this elapsed time is the instantaneous inter-arrival time of the incoming stream. The smoothed inter-arrival time τ is computed using a smoothing factor α (α ≥ 0.5):
( α )
τα
τ = + t 1 −
The meter then passes its information to the scheduler.
When processing a packet, the scheduler attempts to bundle it with other in-queue packets and forward the encapsulating packet as soon as possible. If the queue is empty, the scheduler predicts a potential packet bundling, in a threshold-based manner, by using meter’s information and wireless path characteristics. Details on how to compute the threshold using wireless path characteristics require further research and are out of scope of this paper. If potential packet bundling is not allowed, the packet is encapsulated and forwarded immediately; otherwise, the scheduler injects some small waiting delay to the packet without increasing its end-to-end delay. Upon next arrival or time out, the incoming packet is encapsulated and forwarded respectively with or without packet bundling.
E. Simplification of Tunnel Management
Unlike IP tunneling, virtual SCTP tunneling supports multiple addresses for each endpoint; therefore it reduces the tunnel management complexity and the system resource usage.
Let m denote the number of source endpoint addresses and n denote the number of destination endpoint addresses; it is essential to have only one tunnel device for the communication between the two multi-homed endpoints with virtual SCTP tunneling but mn tunnel devices with IP tunneling. At the Home Agent and Mobile Routers, the destination address list of the virtual SCTP association is synchronized with the Care-of Address list when the Mobile Router is moving by using predefined SCTP primitives such as Add IP Address and Delete IP Address.
IV. PERFORMANCE EVALUATION
A. Simulation Description
We carry out the simulation under network simulator 2.29 with NO Ad-Hoc (NOAH) routing extension and with our extensions for multi-interface Mobile Routers, virtual SCTP
tunneling and manual routing in the hierarchical addressing mode. The scheduler inside the tunnel’s encapsulator distributes the traffic to different Care-of Addresses, i.e. to different radio interfaces, on a per-packet basis, and in a simple manner using the Round-robin algorithm. The meter uses a smoothing factor αof 0.6 for computing the smoothed inter- arrival time.
The IPv6 simulation topology (see Fig. 3) includes one Home Agent and two Access Routers (ARs). Both Access Routers have the capacity of 5.5 Mb/s. Different values of delay are used for different Home Agent-Access Router links as the first step to simulate the difference in wireless path characteristics. Different error transmission characteristics will be considered in our future work. It is assumed that each Mobile Network has only one Mobile Router which has two egress interfaces; each interface associates to one Access Router via the wireless link.
All Mobile Routers are in the coverage of both Access Routers and uniformly positioned in this region. This simulates a fully overlapped coverage area of different radio access technologies and allows Mobile Routers to have simultaneous access to the routing infrastructure, i.e. to the Internet. We consider 5 different Mobile Networks; the number of Local Fixed Nodes inside each Mobile Network is 15, 10, 10, 5, and 5 respectively.
We use two traffic classes to survey service differentiation:
the first traffic class, using G.728 as Pulse Code Modulation codec, consists of 20 VoIP flows, each flow creates VoIP packets of 100 bytes (52-bytes UDP payload) at a 20-ms sample interval; the second traffic class consists of 25 video or data flows, each flow creates packets of 1024 bytes at a 20-ms interval. A random bijection between the set of Correspondent Nodes and the set of Local Fixed Nodes is initialized. A flow is initialized for each pair CN-LFN and is randomly bound to a radio interfaces as follows: for each run, generate a random number r in the range of 0.2 and 0.8 uniformly; for each flow, generate a random number x in the rage of 0 and 1; if x < r, bind the flow to the first interface, otherwise, bind the flow to the second interface. 20 flows of the first traffic class are gradually initialized at random starting time. Later, 25 remaining flows of the second traffic class are initialized in the
Figure 3. Simulation topology
same manner. The observation is carried out, with 20 simulation runs, on the following metrics: system offered load, traffic class throughput and average end-to-end delay. For each metric, we also estimate and plot the 95% confidence interval of sample means.
B. Performance Results and Analysis
Fig. 4 shows results for the first traffic class, and Fig. 5 shows results for the second traffic class. Fig. 4a and 5a represent the impact of system offered load on each traffic class throughput while Fig. 4b, 4c, 5b and 5c represent the impact of number of active Local Fixed Nodes on the packet loss rate and the average end-to-end delay for each traffic class.
In all cases, the virtual SCTP tunneling with predictive packets bundling feature (i.e. the number of encapsulating chunk k is greater than one) provides the highest throughput and the lowest packet loss rate. For example, see Fig. 4b, if the maximum tolerated packet loss rate for a VoIP connection is
5%, the maximum number of supported active Local Fixed Nodes when using IPv6 tunneling, virtual SCTP tunneling without and with bundling is respectively 24 (the worst), 28 (better) and 29 (the best). In heavy-load condition, the virtual SCTP tunneling provides smaller average end-to-end delay and better throughput for both traffic classes thanks to the load- balancing feature. In overload condition, it provides higher throughput and smaller packet loss rate with larger delay as a trade-off. However this trade-off is worth for applications, e.g.
data transfer or video streaming, where lost packets cause negative impact on the QoS. Besides, our observation also shows that service differentiation in NEMO is still an open challenge that requires more research efforts.
V. CONCLUSIONS
This paper proposes a novel intelligent tunneling framework, referred to as virtual SCTP tunneling that enables the Always Best Connected vision in Network Mobility. The
0 1 2 3 4 5 6 7 8 9 10 11
0 0.2 0.4 0.6 0.8 1
Offered load (Mb/s)
Throughput (Mb/s)
a. Througput (class 1)
IPv6 vSCTP k=1 vSCTP k>1
Heavy-load Overload
0 5 10 15 20 25 30 35 40 45
0 5 10 15 20 25 30
Number of active LFNs
Rate (%)
b. Packet loss rate (class 1)
IPv6 vSCTP k=1 vSCTP k>1
0 5 10 15 20 25 30 35 40 45
20 40 60 80 100 120 140 160
Number of active LFNs
Delay (ms)
c. Average end-to-end delay (class 1)
IPv6 vSCTP k=1 vSCTP k>1
Heavy-load Overload
Figure 4. Simulation results for the second traffic class
0 1 2 3 4 5 6 7 8 9 10 11
0 1 2 3 4 5 6
Offered load (Mb/s)
Throughput (Mb/s)
a. Througput (class 2)
IPv6 vSCTP k=1 vSCTP k>1
Heavy-load Overload
0 5 10 15 20 25 30 35 40 45
0 5 10 15 20 25 30
Number of active LFNs
Rate (%)
b. Packet loss rate (class 2)
IPv6 vSCTP k=1 vSCTP k>1
0 5 10 15 20 25 30 35 40 45
20 40 60 80 100 120 140 160
Number of active LFNs
Delay (ms)
c. Average end-to-end delay (class 2)
IPv6 vSCTP k=1 vSCTP k>1
Heavy-load Overload
Figure 5. Simulation results for the second traffic class
new tunneling method reduces the tunneling overhead on radio interfaces and can allow for wireless bandwidth aggregation and per-packet load-balancing to multi-interface Mobile Routers by distributing packets simultaneously via different active radio interfaces on a per-packet basis. Our framework is advantageous over the existing per-flow solution using IP tunneling in terms of throughput, packet loss rate, as well as end-to-end delay. In the future, we will optimize the scheduling algorithm in consideration of mobility, service differentiation and feedback information from Access Routers about the wireless link characteristics and wireless link capacity.
ACKNOWLEDGMENT
Authors are thankful to Hirokazu Naoe (SHARP Corporation) and Lamia Romdhani (Eurecom Institute) for their help and valuable advices.
REFERENCES
[1] E. Gustafsson, and A. Jonsson, "Always best connected", IEEE Wireless communications, Vol. 10, No. 1, pp. 49-55, February 2003.
[2] T. Ernst, “Network Mobility Support Goals and Requirements,” RFC 4886, July 2007.
[3] T. Ernst, H-Y. Lach, “Network Mobility Support Terminology,” RFC 4885, July 2007.
[4] V. Devarapalli, R. Wakikawa, A. Petrescu, and P. Thubert, “Network Mobility (NEMO) Basic Support Protocol,” RFC 3963, January 2005.
[5] D. Johnson , C. Perkins and J. Arkko, “Mobility Support in IPv6,” IETF RFC3775, Jun 2004.
[6] C. Ng, F. Zhao, M. Watari, P. Thubert, “Network Mobility Route Optimization Solution Space Analysis,” RFC 4889, July 2007
[7] R. Wakikawa, T. Ernst, K. Nagami, and V. Devarapalli, “Multiple Care- of Addresses Registration,” Internet draft, draft-ietf-monami6- multiplecoa-03.txt (work in progress), July 2007.
[8] C. Ng, T. Ernst, E. Paik and M. Bagnulo, “Analysis of Multihoming in Network Mobility Support,” Internet draft, draft-ietf-nemo- multihoming-issues-07.txt (work in progress), February 2007.
[9] H. Naoe, M. Wetterwald and C. Bonnet, “IPv6 Soft Handover Applied to Network Mobility over Heterogeneous Access Networks,” PIMRC 2007, 18th IEEE Annual International Symposium on Personal Indoor and Mobile Radio Communications, September 3-7, 2007, Athens, Greece.
[10] Monami6, “Mobile Nodes and multiple interfaces in ipv6,”
https://labo4g.enstb.fr/twiki/bin/view/Monami6/WebHome.
[11] H. Soliman, N. Montavont, N. Fikouras and K. Kuladinithi, « Flow Bindings in Mobile IPv6 and Nemo Basic Support,” Internet draft, draft- soliman-monami6-flow-binding-04.txt (work in progress), February 2007.
[12] H. N. Nguyen and C. Bonnet, “Enhancements for simultaneous access in network-based localized mobility management,” PIMRC 2007, 18th IEEE Annual International Symposium on Personal Indoor and Mobile Radio Communications, September 3-7, 2007, Athens, Greece
[13] R. Stewart, Q. Xie, K. Morneault, H. Schwarzbauer, T. Taylor, I. Rytina, M. Kalla, L. Zhang and V. Paxson, “Stream Control Transmission Protocol,” RFC 2960, October 2000.
[14] S. Fu and M. Atiquzzaman, “Sctp: state of the art in research, products, and technical challenges,” Communications Magazine, IEEE, vol. 42, no. 4, pp. 64–76, 2004.
[15] S.-J. Koh and S.-W. Kim, “msctp for vertical handover between heterogeneous networks,” Web and Communication Technologies and Internet Related Social Issues HSI 2005, 2005.
[16] A. A. E. Al, T. N. Saadawi, and M. J. Lee, “Ls-sctp: a bandwidth aggregation technique for stream control transmission protocol,”
Computer Communications, vol. 27, no. 10, pp. 1012–1024, 2004.
